Well, greetings people of the internet—Uncle Joe here. In this one I’m making the case for “right tool for the right job,” and yeah… sometimes that tool is Windows. I dragged my Dell R730 back into service (dual Xeon E5-2697 v4, 256GB RAM, and dual Tesla P40s with 24GB VRAM each) because I got tired of having good gear just sitting there. I’m not trying to slice these GPUs up for a bunch of VMs anymore—I want to use them natively for stuff like local LLM work, and Windows Server makes that dead simple compared to GPU passthrough gymnastics. I show LM Studio seeing both P40s (in WDDM mode with an older-but-working driver) and splitting a 20B model across both cards, giving me very usable performance—often 20–40+ tokens/sec depending on the prompt. Then I load a 70B model just to prove the point: it’ll run, but you’re trading speed for size (I saw ~3–4 tokens/sec). From there I pivot into why Hyper-V is still one of my favorite “boring and stable” tools for lab VMs, plus a quick look at how I’m doing Windows storage pools (spinning-rust RAID10-ish mirror and a simple SSD pool). Bottom line: Linux is great, Proxmox is great—but when the workload is native Windows + GPUs + Hyper-V, Windows Server is absolutely the right tool.